{"id":"W4321600738","doi":"10.1017/s0261444823000034","title":"The ethical turn in writing assessment: How far have we come, and where do we still need to go?","year":2023,"lang":"en","type":"article","venue":"Language Teaching","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Lethbridge","funders":"","keywords":"Impromptu; Writing assessment; Globe; German; Test (biology); Language assessment; Second language writing; Academic writing; Psychology; Pedagogy; Sociology; Linguistics; Computer science; Second language; Philosophy","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003406753,0.0001373798,0.0001847879,0.0001128123,0.0008326645,0.000755736,0.0003660187,0.0001604989,0.00003636591],"category_scores_gemma":[0.0004354157,0.0001068847,0.00004849897,0.0003321387,0.0001018327,0.0001683929,0.000245423,0.001112555,0.0000295727],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000105538,"about_ca_system_score_gemma":0.00004411337,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001412942,"about_ca_topic_score_gemma":0.01233733,"domain_scores_codex":[0.9978718,0.0006134394,0.000186864,0.0002667103,0.0005450058,0.0005161919],"domain_scores_gemma":[0.9985132,0.001106507,0.00006566571,0.0001748858,0.00001899235,0.0001207186],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00001281363,0.00003471462,0.2349308,0.0000313694,0.00003448808,0.0001063861,0.5493307,0.00001053573,0.001106939,0.04456241,0.004871735,0.164967],"study_design_scores_gemma":[0.0005457467,0.00003900767,0.06445351,0.0002458083,0.00001546861,0.000001245646,0.7846465,0.0008096583,0.000005191363,0.0008719228,0.1480643,0.000301671],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8680614,0.001904041,0.0001100693,0.1038082,0.0002749028,0.0004266483,0.000006757708,0.0002388282,0.02516912],"genre_scores_gemma":[0.9924238,0.0007146916,0.0006525669,0.0003321441,0.0002729509,0.00002353945,0.000002915496,0.00001911481,0.005558273],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2353157,"threshold_uncertainty_score":0.728758,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02481394901592722,"score_gpt":0.3791913870123313,"score_spread":0.3543774379964041,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}